Termination Conditions

Termination conditions stop a multi-agent loop before it hits maxTurns. They are particularly useful in swarm and conversation patterns where you don't know upfront how many turns are needed.

Available conditions

MaxMessageTermination

Stop after a fixed number of messages:

import org.conductoross.conductor.ai.termination.MaxMessageTermination;

Agent agent = Agent.builder()
    .termination(MaxMessageTermination.of(10))
    .build();

StopMessageTermination

Stop when the agent outputs a specific string:

import org.conductoross.conductor.ai.termination.StopMessageTermination;

Agent agent = Agent.builder()
    .termination(StopMessageTermination.of("TASK_COMPLETE"))
    .build();

Instruct the agent in its system prompt:

When you have finished the task, output exactly: TASK_COMPLETE

TextMentionTermination

Stop when the output mentions specific text (optionally case-sensitive):

import org.conductoross.conductor.ai.termination.TextMentionTermination;

// Case-insensitive (default)
TextMentionTermination.of("done")

// Case-sensitive
TextMentionTermination.of("DONE", true)

TokenUsageTermination

Stop when cumulative token usage exceeds a limit — useful for cost control:

import org.conductoross.conductor.ai.termination.TokenUsageTermination;

// Stop after 50,000 total tokens
Agent agent = Agent.builder()
    .termination(TokenUsageTermination.ofTotal(50_000))
    .build();

Composing conditions

Chain conditions with and() / or():

// Stop when BOTH are true: max 20 messages AND output contains "DONE"
TerminationCondition both = MaxMessageTermination.of(20)
    .and(StopMessageTermination.of("DONE"));

// Stop when EITHER is true: max 20 messages OR output contains "DONE"
TerminationCondition either = MaxMessageTermination.of(20)
    .or(StopMessageTermination.of("DONE"));

Agent agent = Agent.builder()
    .termination(either)
    .build();

Stopping a running execution

Termination conditions are part of an agent definition. Operator-driven execution control uses AgentClient instead:

// Graceful: finish the current iteration, then stop.
agentClient.stopAgent(executionId);

// Immediate: terminate now and record an optional reason.
agentClient.cancelAgent(executionId, "Superseded by a newer run");

stopAgent sends POST /agent/{id}/stop. cancelAgent sends DELETE /agent/{id}/cancel and omits the reason query parameter when it is null or blank. See the AgentClient reference.